Pub Date : 2025-12-11DOI: 10.1016/j.phro.2025.100890
Mathieu Gaudreault , Lachlan McIntosh , Katrina Woodford , Jason Li , Susan Harden , Sandro Porceddu , Nicholas Hardcastle , Vanessa Panettieri
Background and purpose
The monitor units (MU) per control point (CP) control the necessary fine-tuned ablative dose for hypofractionated radiotherapy of oligometastatic cancer. We aimed to introduce strategies maximising the sample size to accurately predict the MU per CP with artificial intelligence (AI).
Materials and methods
The 40/68/88 treatment plans of consecutive patients treated between 01/2019 and 06/2024 at our institution for metastatic cancer to the liver/bone/lung were included. Two approaches were considered to maximise the sample size. In one approach, the samples of each anatomical site were extensively augmented to predict the MU per CP from the dose distribution per CP, providing the MU per beam and meterset weight per CP. In the other approach, all samples from all anatomical sites were combined for training. The number of achieved clinical goals based on dose-volume calculation metrics in AI radiotherapy plans (AI-RTPlan) was compared with the number of achieved clinical goals in the clinical plans.
Results
The mean absolute percentage error between predicted and clinical MU per beam/meterset weight per CP was less than 6.2%. All AI-RTPlans were generated in less than 5 s. At least 90%/5% of patients had the same, or more, achieved clinical goals with AI-RTPlans. Target coverage and dose to organs at risk metrics were within ± 2% and ± 2.3 Gy of the clinical value in all patients, respectively.
Conclusions
Augmenting data extensively and combining anatomical sites were equivalent and proficient strategies to predict machine settings for radiotherapy planning of oligometastatic cancer.
{"title":"A deep learning approach for predicting linear accelerator output settings in automated radiotherapy planning of oligometastatic cancer","authors":"Mathieu Gaudreault , Lachlan McIntosh , Katrina Woodford , Jason Li , Susan Harden , Sandro Porceddu , Nicholas Hardcastle , Vanessa Panettieri","doi":"10.1016/j.phro.2025.100890","DOIUrl":"10.1016/j.phro.2025.100890","url":null,"abstract":"<div><h3>Background and purpose</h3><div>The monitor units (MU) per control point (CP) control the necessary fine-tuned ablative dose for hypofractionated radiotherapy of oligometastatic cancer. We aimed to introduce strategies maximising the sample size to accurately predict the MU per CP with artificial intelligence (AI).</div></div><div><h3>Materials and methods</h3><div>The 40/68/88 treatment plans of consecutive patients treated between 01/2019 and 06/2024 at our institution for metastatic cancer to the liver/bone/lung were included. Two approaches were considered to maximise the sample size. In one approach, the samples of each anatomical site were extensively augmented to predict the MU per CP from the dose distribution per CP, providing the MU per beam and meterset weight per CP. In the other approach, all samples from all anatomical sites were combined for training. The number of achieved clinical goals based on dose-volume calculation metrics in AI radiotherapy plans (AI-RTPlan) was compared with the number of achieved clinical goals in the clinical plans.</div></div><div><h3>Results</h3><div>The mean absolute percentage error between predicted and clinical MU per beam/meterset weight per CP was less than 6.2%. All AI-RTPlans were generated in less than 5 s. At least 90%/5% of patients had the same, or more, achieved clinical goals with AI-RTPlans. Target coverage and dose to organs at risk metrics were within ± 2% and ± 2.3 Gy of the clinical value in all patients, respectively.</div></div><div><h3>Conclusions</h3><div>Augmenting data extensively and combining anatomical sites were equivalent and proficient strategies to predict machine settings for radiotherapy planning of oligometastatic cancer.</div></div>","PeriodicalId":36850,"journal":{"name":"Physics and Imaging in Radiation Oncology","volume":"37 ","pages":"Article 100890"},"PeriodicalIF":3.3,"publicationDate":"2025-12-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145760658","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-10-01DOI: 10.1016/j.phro.2025.100880
Anne L.H. Bisgaard , Chavelli M. Kensen , Marielle E.P. Philippens , Martijn P.W. Intven , Gert J. Meijer , Alice M. Couwenberg , Doenja M.J. Lambregts , Uulke A. van der Heide , Erik van der Bijl , Pètra M. Braam , Faisal Mahmood , Petra J. van Houdt
Background and purpose
The apparent diffusion coefficient (ADC) derived from diffusion-weighted imaging (DWI), a form of magnetic resonance imaging (MRI), has shown promise for predicting response to long course neoadjuvant chemoradiotherapy in rectal cancer. This study investigated whether ADC changes are detectable during short course radiotherapy in patients with rectal cancer.
Materials and methods
Across 3 centres, this study included 138 patients with primary tumours, who received neoadjuvant short course radiotherapy (5 fractions of 5 Gy) on a 1.5 T MRI linear accelerator (MRI-linac), without any prior oncological treatments. DWI was acquired at each fraction prior to beam-on. ADC maps were calculated centrally using a mono-exponential model using b-values between 150–800 s/mm2. Median scaling of ADC voxel values was performed between two identified groups of DWI sequences. Tumours were semi-automatically delineated on DWI, and median ADCs were extracted per fraction. ADC time-trends over the course of radiotherapy were extracted using linear fitting, with 95% confidence intervals (CI) estimated using bootstrapping.
Results
A scaling factor of 0.93 was used to account for ADC variation between the DWI sequence groups. The median (range) slope of the ADC time-trends was 0.05 (−0.18, 0.42) 10−3mm2/s/fraction. In 77 patients (56%), the 95% CI of the slope did not include zero.
Conclusions
ADC changes during short course radiotherapy were detectable in 56% of the patients. Furthermore, the limited ADC variation across DWI sequences supports feasibility of multicentre investigations of MRI-linac based DWI. These findings encourage future research linking ADC to clinical outcomes in rectal cancer for potential treatment personalization.
背景与目的磁共振成像（MRI）的一种形式——弥散加权成像（DWI）得出的表观扩散系数（ADC）有望预测直肠癌患者对长期新辅助放化疗的反应。本研究探讨了在直肠癌患者的短期放疗中是否可以检测到ADC的变化。材料和方法本研究纳入了3个中心的138例原发肿瘤患者，这些患者在1.5 T MRI直线加速器（MRI-linac）上接受了新辅助短期放疗（5 Gy的5个部分），之前没有任何肿瘤治疗。在光束照射前，在每个分数处获取DWI。ADC图使用单指数模型集中计算，b值在150-800 s/mm2之间。在确定的两组DWI序列之间进行ADC体素值的中位数缩放。在DWI上半自动划定肿瘤，并提取每个分数的中位adc。放疗过程中的ADC时间趋势采用线性拟合提取，95%置信区间（CI）采用自举法估计。结果DWI序列组间ADC差异的比例因子为0.93。ADC时间趋势的中位（范围）斜率为0.05 (- 0.18,0.42)10 - 3mm2/s/fraction。在77例（56%）患者中，斜率的95% CI不为零。结论56%的患者在短期放疗中可检测到sadc的改变。此外，DWI序列之间有限的ADC变化支持了基于MRI-linac的DWI多中心研究的可行性。这些发现鼓励未来的研究将ADC与直肠癌的临床结果联系起来，以实现潜在的个性化治疗。
{"title":"Apparent diffusion coefficient increases during short course radiotherapy in rectal tumours: Results from a multicentre longitudinal trial","authors":"Anne L.H. Bisgaard , Chavelli M. Kensen , Marielle E.P. Philippens , Martijn P.W. Intven , Gert J. Meijer , Alice M. Couwenberg , Doenja M.J. Lambregts , Uulke A. van der Heide , Erik van der Bijl , Pètra M. Braam , Faisal Mahmood , Petra J. van Houdt","doi":"10.1016/j.phro.2025.100880","DOIUrl":"10.1016/j.phro.2025.100880","url":null,"abstract":"<div><h3>Background and purpose</h3><div>The apparent diffusion coefficient (ADC) derived from diffusion-weighted imaging (DWI), a form of magnetic resonance imaging (MRI), has shown promise for predicting response to long course neoadjuvant chemoradiotherapy in rectal cancer. This study investigated whether ADC changes are detectable during short course radiotherapy in patients with rectal cancer.</div></div><div><h3>Materials and methods</h3><div>Across 3 centres, this study included 138 patients with primary tumours, who received neoadjuvant short course radiotherapy (5 fractions of 5 Gy) on a 1.5 T MRI linear accelerator (MRI-linac), without any prior oncological treatments. DWI was acquired at each fraction prior to beam-on. ADC maps were calculated centrally using a mono-exponential model using b-values between 150–800 s/mm<sup>2</sup>. Median scaling of ADC voxel values was performed between two identified groups of DWI sequences. Tumours were semi-automatically delineated on DWI, and median ADCs were extracted per fraction. ADC time-trends over the course of radiotherapy were extracted using linear fitting, with 95% confidence intervals (CI) estimated using bootstrapping.</div></div><div><h3>Results</h3><div>A scaling factor of 0.93 was used to account for ADC variation between the DWI sequence groups. The median (range) slope of the ADC time-trends was 0.05 (−0.18, 0.42) 10<sup>−3</sup>mm<sup>2</sup>/s/fraction. In 77 patients (56%), the 95% CI of the slope did not include zero.</div></div><div><h3>Conclusions</h3><div>ADC changes during short course radiotherapy were detectable in 56% of the patients. Furthermore, the limited ADC variation across DWI sequences supports feasibility of multicentre investigations of MRI-linac based DWI. These findings encourage future research linking ADC to clinical outcomes in rectal cancer for potential treatment personalization.</div></div>","PeriodicalId":36850,"journal":{"name":"Physics and Imaging in Radiation Oncology","volume":"36 ","pages":"Article 100880"},"PeriodicalIF":3.3,"publicationDate":"2025-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145681028","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-10-01DOI: 10.1016/j.phro.2025.100879
José Antonio Baeza-Ortega , Lauren May , Mohammad Hussein , Sarah Porter , Alisha Moore , Peter B. Greer , Catharine H. Clark , Joerg Lehmann
Background and purpose
The role of dosimetry audits is well established in the development and verification of radiotherapy safety. Differences in planning and beam modelling make inter-centre comparisons challenging, which can be addressed through distribution of centrally created plans. This study developed a centralised planning approach applicable to multiple audit methodologies, using an example of remote patient specific quality assurance assessment, increasing the interpretability of results and facilitating automation and scalability.
Material and methods
Starting with an established plan which met all clinical goals, a commercial dose mimicking algorithm was used to replicate this plan to be suitable for multiple treatment machines. Beam and machine limitation data were collected from participating centres to develop universally acceptable beam models. The influence of variation in beam modelling parameters among centres was assessed by creating additional models using the 2.5th, 25th, 75th and 97.5th percentiles of previously reported data. Multi-leaf collimator angle and leaf position, gantry angle and output deviations were then introduced into copies of these plans.
Results
Introduced delivery errors caused consistent change in dose metrics across machine models (excluding outliers) with a median (range) standard deviation of 1.0 % (from 0.1 % to 1.7 %) demonstrating similar robustness. Beam model variation did not change whether simulated delivery errors were clinically impactful or not for 95 % of tested plans.
Conclusion
This study lays the foundation for future standardised methodology for dosimetry audits by providing a centralised planning approach that allows a more consistent assessment of centres.
{"title":"A proof of concept for improving comparability of dosimetry audits through centralised planning","authors":"José Antonio Baeza-Ortega , Lauren May , Mohammad Hussein , Sarah Porter , Alisha Moore , Peter B. Greer , Catharine H. Clark , Joerg Lehmann","doi":"10.1016/j.phro.2025.100879","DOIUrl":"10.1016/j.phro.2025.100879","url":null,"abstract":"<div><h3>Background and purpose</h3><div>The role of dosimetry audits is well established in the development and verification of radiotherapy safety. Differences in planning and beam modelling make inter-centre comparisons challenging, which can be addressed through distribution of centrally created plans. This study developed a centralised planning approach applicable to multiple audit methodologies, using an example of remote patient specific quality assurance assessment, increasing the interpretability of results and facilitating automation and scalability.</div></div><div><h3>Material and methods</h3><div>Starting with an established plan which met all clinical goals, a commercial dose mimicking algorithm was used to replicate this plan to be suitable for multiple treatment machines. Beam and machine limitation data were collected from participating centres to develop universally acceptable beam models. The influence of variation in beam modelling parameters among centres was assessed by creating additional models using the 2.5th, 25th, 75th and 97.5th percentiles of previously reported data. Multi-leaf collimator angle and leaf position, gantry angle and output deviations were then introduced into copies of these plans.</div></div><div><h3>Results</h3><div>Introduced delivery errors caused consistent change in dose metrics across machine models (excluding outliers) with a median (range) standard deviation of 1.0 % (from 0.1 % to 1.7 %) demonstrating similar robustness. Beam model variation did not change whether simulated delivery errors were clinically impactful or not for 95 % of tested plans.</div></div><div><h3>Conclusion</h3><div>This study lays the foundation for future standardised methodology for dosimetry audits by providing a centralised planning approach that allows a more consistent assessment of centres.</div></div>","PeriodicalId":36850,"journal":{"name":"Physics and Imaging in Radiation Oncology","volume":"36 ","pages":"Article 100879"},"PeriodicalIF":3.3,"publicationDate":"2025-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145681029","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-10-01DOI: 10.1016/j.phro.2025.100862
Chunbo Tang , Houjin Zhang , Longqiu Wu , Minfeng Huang , Pengfei Wang , Jun Yuan , Junjie Zhang , Biaoshui Liu , Ji He
Hippocampal avoidance whole-brain radiotherapy (HA-WBRT) aims to preserve cognitive function during treatment for brain metastases. This study investigated the potential of Single-Target Longitudinal Segmentation Volumetric Modulated Arc Therapy (VMAT) in HA-WBRT, which segments the planning target volume (PTV) into sub-PTVs, using single or dual arcs to generate s-VMAT and d-VMAT strategies. For 20 patients, s-VMAT and d-VMAT achieved lower median Dmean values of 8.3 Gy and 8.1 Gy, and reduced the median Dmax to 13.5 Gy and 12.8 Gy, compared to traditional coplanar/non-coplanar VMAT plans. These strategies showed enhanced robustness but required more monitor units and greater delivery complexity.
{"title":"Positioning uncertainties in single-target longitudinal segmentation for hippocampal-avoidance whole brain radiotherapy using volumetric modulated arc therapy","authors":"Chunbo Tang , Houjin Zhang , Longqiu Wu , Minfeng Huang , Pengfei Wang , Jun Yuan , Junjie Zhang , Biaoshui Liu , Ji He","doi":"10.1016/j.phro.2025.100862","DOIUrl":"10.1016/j.phro.2025.100862","url":null,"abstract":"<div><div>Hippocampal avoidance whole-brain radiotherapy (HA-WBRT) aims to preserve cognitive function during treatment for brain metastases. This study investigated the potential of Single-Target Longitudinal Segmentation Volumetric Modulated Arc Therapy (VMAT) in HA-WBRT, which segments the planning target volume (PTV) into sub-PTVs, using single or dual arcs to generate s-VMAT and d-VMAT strategies. For 20 patients, s-VMAT and d-VMAT achieved lower median Dmean values of 8.3 Gy and 8.1 Gy, and reduced the median Dmax to 13.5 Gy and 12.8 Gy, compared to traditional coplanar/non-coplanar VMAT plans. These strategies showed enhanced robustness but required more monitor units and greater delivery complexity.</div></div>","PeriodicalId":36850,"journal":{"name":"Physics and Imaging in Radiation Oncology","volume":"36 ","pages":"Article 100862"},"PeriodicalIF":3.3,"publicationDate":"2025-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145681032","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-10-01DOI: 10.1016/j.phro.2025.100886
Libing Zhu , Yi Rong , Nathan Y. Yu , Jason M. Holmes , Carlos E. Vargas , Sarah E. James , Lu Shang , Jean-Claude M. Rwigema , Quan Chen
Background and purpose
Deep-learning auto-segmentation (DLAS) performance in radiotherapy may change over time due to data shift/drift or practice changes, yet guidance for quality assurance is lacking. This study developed a practical framework for prospective performance monitoring using retrospective data.
Methods
A total of 464 prostate cases over 20 months were retrospectively collected. Two commercial DLAS models were clinically used: model A (2D U-Net, January 2022–January 2023) and model B (3D U-Net, February–August 2023). The agreement between DLAS and clinical contours was assessed using Dice Similarity Coefficient (DSC), 95th percentile Hausdorff Distance (HD95), and Surface DSC with a 2 mm tolerance (SDSC). Statistical process control charts were created to monitor performance drift and model switching. The first 150 cases were used to define organ-specific control limits with two and three standard deviations of monthly mean values, .
Results
2 and 3-based control limits were established for the monthly average charts, ranging from DSC 0.82–0.97, HD95 1.4–10.5 mm, and SDSC 0.45–0.91 across organs. Model A showed stable performance, with 9–13 months per organ remaining within the 3 thresholds. In contrast, model B demonstrated a marked performance shift (p < 0.001), with all five organs exceeding both thresholds across all 7 months. The 2 thresholds were more sensitive in detecting mild deviations for model A, while both limits effectively identified the substantial drift of model B.
Conclusion
The monitoring system effectively detected out-of-distribution outliers and clinical practice changes, providing a reliable framework for early detection of monthly performance degradation.
{"title":"Establishing prospective performance monitoring for real-world implementation of deep learning-based auto-segmentation in prostate cancer radiotherapy","authors":"Libing Zhu , Yi Rong , Nathan Y. Yu , Jason M. Holmes , Carlos E. Vargas , Sarah E. James , Lu Shang , Jean-Claude M. Rwigema , Quan Chen","doi":"10.1016/j.phro.2025.100886","DOIUrl":"10.1016/j.phro.2025.100886","url":null,"abstract":"<div><h3>Background and purpose</h3><div>Deep-learning auto-segmentation (DLAS) performance in radiotherapy may change over time due to data shift/drift or practice changes, yet guidance for quality assurance is lacking. This study developed a practical framework for prospective performance monitoring using retrospective data.</div></div><div><h3>Methods</h3><div>A total of 464 prostate cases over 20 months were retrospectively collected. Two commercial DLAS models were clinically used: model A (2D U-Net, January 2022–January 2023) and model B (3D U-Net, February–August 2023). The agreement between DLAS and clinical contours was assessed using Dice Similarity Coefficient (DSC), 95th percentile Hausdorff Distance (HD95), and Surface DSC with a 2 mm tolerance (SDSC). Statistical process control charts were created to monitor performance drift and model switching. The first 150 cases were used to define organ-specific control limits with two and three standard deviations of monthly mean values, <span><math><mrow><msub><mi>σ</mi><mover><mrow><mi>x</mi></mrow><mrow><mo>¯</mo></mrow></mover></msub></mrow></math></span>.</div></div><div><h3>Results</h3><div>2<span><math><mrow><msub><mi>σ</mi><mover><mrow><mi>x</mi></mrow><mrow><mo>¯</mo></mrow></mover></msub></mrow></math></span> and 3<span><math><mrow><msub><mi>σ</mi><mover><mrow><mi>x</mi></mrow><mrow><mo>¯</mo></mrow></mover></msub></mrow></math></span>-based control limits were established for the monthly average charts, ranging from DSC 0.82–0.97, HD95 1.4–10.5 mm, and SDSC 0.45–0.91 across organs. Model A showed stable performance, with 9–13 months per organ remaining within the 3<span><math><mrow><msub><mi>σ</mi><mover><mrow><mi>x</mi></mrow><mrow><mo>¯</mo></mrow></mover></msub></mrow></math></span> thresholds. In contrast, model B demonstrated a marked performance shift (p < 0.001), with all five organs exceeding both thresholds across all 7 months. The 2<span><math><mrow><msub><mi>σ</mi><mover><mrow><mi>x</mi></mrow><mrow><mo>¯</mo></mrow></mover></msub></mrow></math></span> thresholds were more sensitive in detecting mild deviations for model A, while both limits effectively identified the substantial drift of model B.</div></div><div><h3>Conclusion</h3><div>The monitoring system effectively detected out-of-distribution outliers and clinical practice changes, providing a reliable framework for early detection of monthly performance degradation.</div></div>","PeriodicalId":36850,"journal":{"name":"Physics and Imaging in Radiation Oncology","volume":"36 ","pages":"Article 100886"},"PeriodicalIF":3.3,"publicationDate":"2025-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145736394","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Oligometastatic disease represents limited metastatic burden, and local ablative therapies such as stereotactic body radiotherapy (SBRT) may improve survival. However, inter-institutional variability in target segmentation and treatment planning can compromise treatment quality. This study aimed to evaluate the segmentation variability and dose distribution quality of SBRT in oligometastatic settings using a multi-institutional dummy run approach.
Methods and materials
Sixty-nine institutions were provided with two anonymized cases of adrenal and spine metastases to delineate targets and organs at risk (OARs) and create intensity-modulated radiotherapy plans following a protocol. Variability was quantified using the Dice similarity coefficient (DSC), Hausdorff distance, and mean distance to agreement. Plan qualities were assessed using the Paddick conformity index, modified gradient index, and a new three-dimensional conformity–gradient index (3D-CGI). Knowledge-based planning (KBP) was applied to explore potential improvements in OAR sparing.
Results
All submitted plans met protocol dose constraints. However, substantial segmentation variability was observed, particularly for the spine case. Among 136 plans, 79% demonstrated acceptable conformity and dose gradients, with 3D-CGI < 6 correlating with favorable distributions. Mean DSC was 0.93 for the clinical target volume and 0.76 for the cauda equina, which showed the highest variability. KBP reduced OAR doses for the adrenal case but showed limited impact for the spine case.
Conclusions
Although dose constraints were achieved, segmentation variability remained substantial, particularly for the cauda equina in the spine case. These findings emphasize inter-institutional differences and the need for standardization and tools to improve SBRT consistency.
{"title":"A multi-institutional dummy run on segmentation variability and plan quality of stereotactic body radiotherapy for oligometastatic disease","authors":"Hideaki Hirashima , Yukinori Matsuo , Satoshi Ishikura , Mitsuhiro Nakamura , Ikuno Nishibuchi , Daisuke Kawahara , Yoshihisa Shimada , Yoshiro Nakahara , Teiji Nishio , Naoto Shikama , Shun-ichi Watanabe , Isamu Okamoto , Toshiyuki Ishiba , Fumikata Hara , Tadahiko Shien , Takashi Mizowaki","doi":"10.1016/j.phro.2025.100857","DOIUrl":"10.1016/j.phro.2025.100857","url":null,"abstract":"<div><h3>Background and purpose</h3><div>Oligometastatic disease represents limited metastatic burden, and local ablative therapies such as stereotactic body radiotherapy (SBRT) may improve survival. However, inter-institutional variability in target segmentation and treatment planning can compromise treatment quality. This study aimed to evaluate the segmentation variability and dose distribution quality of SBRT in oligometastatic settings using a multi-institutional dummy run approach.</div></div><div><h3>Methods and materials</h3><div>Sixty-nine institutions were provided with two anonymized cases of adrenal and spine metastases to delineate targets and organs at risk (OARs) and create intensity-modulated radiotherapy plans following a protocol. Variability was quantified using the Dice similarity coefficient (DSC), Hausdorff distance, and mean distance to agreement. Plan qualities were assessed using the Paddick conformity index, modified gradient index, and a new three-dimensional conformity–gradient index (3D-CGI). Knowledge-based planning (KBP) was applied to explore potential improvements in OAR sparing.</div></div><div><h3>Results</h3><div>All submitted plans met protocol dose constraints. However, substantial segmentation variability was observed, particularly for the spine case. Among 136 plans, 79% demonstrated acceptable conformity and dose gradients, with 3D-CGI < 6 correlating with favorable distributions. Mean DSC was 0.93 for the clinical target volume and 0.76 for the cauda equina, which showed the highest variability. KBP reduced OAR doses for the adrenal case but showed limited impact for the spine case.</div></div><div><h3>Conclusions</h3><div>Although dose constraints were achieved, segmentation variability remained substantial, particularly for the cauda equina in the spine case. These findings emphasize inter-institutional differences and the need for standardization and tools to improve SBRT consistency.</div></div>","PeriodicalId":36850,"journal":{"name":"Physics and Imaging in Radiation Oncology","volume":"36 ","pages":"Article 100857"},"PeriodicalIF":3.3,"publicationDate":"2025-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145520095","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-10-01DOI: 10.1016/j.phro.2025.100863
Zhunhao Zheng , Junqiang Chen , Xiaolin Ge , Meng Yan , Ling Li , Qifeng Wang , Xiaomin Wang , Xin Wang , Wenyang Liu , Yonggang Shi , Xiaofeng Wang , Hongyun Shi , Zhilong Yu , Qingsong Pang , Zefen Xiao , Wencheng Zhang
Background and purpose
To integrate clinical characteristics, radiomics, and dosiomics to provide accurate and individualized prediction of radiation pneumonitis (RP) in elderly patients aged 70 and over with esophageal cancer receiving radiotherapy.
Materials and methods
Based on a phase III clinical study (NCT02979691) that included elderly patients with esophageal squamous cell carcinoma (ESCC) who received definitive radiotherapy, we selected a total of 229 patients with available computed tomography (CT) and dose images. Radiomic and dosiomics features were extracted from both lungs. The patients were randomly assigned to either the training group (N = 161) or the test group (N = 68) in a 7:3 ratio. In the training set, logistic regression (LR) was applied to calculate the radiomic score (R score) and dosiomic score (D score). The constructed multivariate LR and ridge regression prediction models were evaluated using the test set. The endpoint of the predictive model is defined as a grade ≥ 2 RP. Discrimination and prediction were assessed by calculating the area under curve (AUC) of the receiver operating characteristic curve and plotting calibration and decision curve analyses (DCA).
Results
The hybrid LR model integrating R score, D score and clinical characteristics had the best clinical applicability. The hybrid model demonstrated superior predictive performance on the test set, achieving an area under the curve (AUC) of 0.76, while the combined clinical and DVH model achieved an AUC of 0.70.
Conclusions
A hybrid model combining radiomics and dosiomics with clinical characteristics showed the best performance for predicting RP.
{"title":"A machine learning approach for radiation pneumonitis prediction in elderly esophageal cancer patients by integrating baseline computed tomography radiomics, dosiomics, and clinical characteristics","authors":"Zhunhao Zheng , Junqiang Chen , Xiaolin Ge , Meng Yan , Ling Li , Qifeng Wang , Xiaomin Wang , Xin Wang , Wenyang Liu , Yonggang Shi , Xiaofeng Wang , Hongyun Shi , Zhilong Yu , Qingsong Pang , Zefen Xiao , Wencheng Zhang","doi":"10.1016/j.phro.2025.100863","DOIUrl":"10.1016/j.phro.2025.100863","url":null,"abstract":"<div><h3>Background and purpose</h3><div>To integrate clinical characteristics, radiomics, and dosiomics to provide accurate and individualized prediction of radiation pneumonitis (RP) in elderly patients aged 70 and over with esophageal cancer receiving radiotherapy.</div></div><div><h3>Materials and methods</h3><div>Based on a phase III clinical study (NCT02979691) that included elderly patients with esophageal squamous cell carcinoma (ESCC) who received definitive radiotherapy, we selected a total of 229 patients with available computed tomography (CT) and dose images. Radiomic and dosiomics features were extracted from both lungs. The patients were randomly assigned to either the training group (N = 161) or the test group (N = 68) in a 7:3 ratio. In the training set, logistic regression (LR) was applied to calculate the radiomic score (R score) and dosiomic score (D score). The constructed multivariate LR and ridge regression prediction models were evaluated using the test set. The endpoint of the predictive model is defined as a grade ≥ 2 RP. Discrimination and prediction were assessed by calculating the area under curve (AUC) of the receiver operating characteristic curve and plotting calibration and decision curve analyses (DCA).</div></div><div><h3>Results</h3><div>The hybrid LR model integrating R score, D score and clinical characteristics had the best clinical applicability. The hybrid model demonstrated superior predictive performance on the test set, achieving an area under the curve (AUC) of 0.76, while the combined clinical and DVH model achieved an AUC of 0.70.</div></div><div><h3>Conclusions</h3><div>A hybrid model combining radiomics and dosiomics with clinical characteristics showed the best performance for predicting RP.</div></div>","PeriodicalId":36850,"journal":{"name":"Physics and Imaging in Radiation Oncology","volume":"36 ","pages":"Article 100863"},"PeriodicalIF":3.3,"publicationDate":"2025-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145576373","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-10-01DOI: 10.1016/j.phro.2025.100877
Mikkel Skaarup, Nikolaj Kylling Gyldenløve Jensen
Background and purpose
The desire to avoid tattooing radiotherapy patients lead us to implement surface-guided initial patient set-up (SGRT). To validate SGRT we investigated setup precision, user attitude, and impact on radiotherapy technician (RTT) workflow.
Materials and methods
During a six-month period, initial setup was investigated on six linear accelerators (Truebeam, Varian), each equipped with a thermo-optical surface camera (ExacTrac Dynamic, Brainlab). Precision was assessed by comparing couch shifts based on x-ray imaging acquired after initial setup and number of x-ray imaging procedures for each fraction to data from the prior year, using a tattoo-based setup. The data was split into subgroups (thoracic, abdominal/pelvic, palliative and miscellaneous (gastrointestinal, head and neck, cranial and extremities)). User attitude and impact on RTT workflow was assessed qualitatively by questionnaire. RTTs were asked to rate how SGRT compared to tattoo-based setup and record the need for manual adjustments of the patient (e.g. pushing, lifting or pulling). Questionnaires were repeated 1.5 years after implementation.
Results
We included 460 patients setup with SGRT, and 468 patients with tattoo-based methods. Median couch shifts and repeated imaging were comparable overall (0.6 cm and 9 % respectively for both setup methods), SGRT performed better for the thoracic and miscellaneous sites subgroups. RTTs preferred SGRT to laser and tattoo initial setup for >90 % of fractions. Manual adjustments were reduced with SGRT (15 % of fractions) compared to tattoo-based (60 % of fractions).
Conclusions
SGRT achieved the same or better precision as tattoo-based initial setup while providing a better workflow and reduced physical adjustments performed by the RTTs by 75 %.
{"title":"Workflow evaluation of surface-guided initial patient set-up in radiotherapy","authors":"Mikkel Skaarup, Nikolaj Kylling Gyldenløve Jensen","doi":"10.1016/j.phro.2025.100877","DOIUrl":"10.1016/j.phro.2025.100877","url":null,"abstract":"<div><h3>Background and purpose</h3><div>The desire to avoid tattooing radiotherapy patients lead us to implement surface-guided initial patient set-up (SGRT). To validate SGRT we investigated setup precision, user attitude, and impact on radiotherapy technician (RTT) workflow.</div></div><div><h3>Materials and methods</h3><div>During a six-month period, initial setup was investigated on six linear accelerators (Truebeam, Varian), each equipped with a thermo-optical surface camera (ExacTrac Dynamic, Brainlab). Precision was assessed by comparing couch shifts based on x-ray imaging acquired after initial setup and number of x-ray imaging procedures for each fraction to data from the prior year, using a tattoo-based setup. The data was split into subgroups (thoracic, abdominal/pelvic, palliative and miscellaneous (gastrointestinal, head and neck, cranial and extremities)). User attitude and impact on RTT workflow was assessed qualitatively by questionnaire. RTTs were asked to rate how SGRT compared to tattoo-based setup and record the need for manual adjustments of the patient (e.g. pushing, lifting or pulling). Questionnaires were repeated 1.5 years after implementation.</div></div><div><h3>Results</h3><div>We included 460 patients setup with SGRT, and 468 patients with tattoo-based methods. Median couch shifts and repeated imaging were comparable overall (0.6 cm and 9 % respectively for both setup methods), SGRT performed better for the thoracic and miscellaneous sites subgroups. RTTs preferred SGRT to laser and tattoo initial setup for >90 % of fractions. Manual adjustments were reduced with SGRT (15 % of fractions) compared to tattoo-based (60 % of fractions).</div></div><div><h3>Conclusions</h3><div>SGRT achieved the same or better precision as tattoo-based initial setup while providing a better workflow and reduced physical adjustments performed by the RTTs by 75 %.</div></div>","PeriodicalId":36850,"journal":{"name":"Physics and Imaging in Radiation Oncology","volume":"36 ","pages":"Article 100877"},"PeriodicalIF":3.3,"publicationDate":"2025-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145680998","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-10-01DOI: 10.1016/j.phro.2025.100852
Arkadiy Dushatskiy , Peter A.N. Bosman , Karel A. Hinnen , Jan Wiersma , Henrike Westerveld , Bradley R. Pieters , Tanja Alderliesten
Background and Purpose
Recently, we introduced a novel Deep Learning (DL) based (semi-)automatic method for medical image segmentation. Unlike classical DL segmentation methods, it produces multiple segmentation variants (reflecting the variation of manual segmentations) instead of just one. Potentially, with this approach, there is a higher chance that a clinician prefers one of automatically produced segmentation variants. This work focuses on evaluating this method on prostate segmentation in MRI scans used for brachytherapy and investigating its potential clinical usefulness.
Materials and Methods
Three experienced radiation oncologists graded (per-slice and per-scan) segmentations produced by our method, reference segmentations (manually created and used for brachytherapy treatment planning) and segmentations produced by a classical DL method. The study was retrospective and the way the segmentation was generated (our method, classical DL method, or manually) was blinded for the clinicians. The grades reflect the amount of manual correction required. Additionally, the clinicians were asked to rank segmentations to evaluate which one is preferred for each scan. The study was performed on 13 prostate cancer patients.
Results
Segmentations produced by our method are graded as requiring no manual correction in 292/576 (51 %) slices compared to 240/576 (42 %) slices in the case of the segmentations produced by a classical DL method. Furthermore, in fewer slices, 38 (6.6 %) vs. 48 (8.3 %), segmentations by our method were graded as unacceptable.
Conclusion
Our study has demonstrated that deep-learning-based segmentation methods can produce high-quality segmentations. Our method was evaluated better than a classical DL method, indicating the potential for integration into clinical practice.
{"title":"Evaluating the quality of multiple automatically produced segmentation variants of the prostate on Magnetic Resonance Imaging scans for brachytherapy","authors":"Arkadiy Dushatskiy , Peter A.N. Bosman , Karel A. Hinnen , Jan Wiersma , Henrike Westerveld , Bradley R. Pieters , Tanja Alderliesten","doi":"10.1016/j.phro.2025.100852","DOIUrl":"10.1016/j.phro.2025.100852","url":null,"abstract":"<div><h3>Background and Purpose</h3><div>Recently, we introduced a novel Deep Learning (DL) based (semi-)automatic method for medical image segmentation. Unlike classical DL segmentation methods, it produces multiple segmentation variants (reflecting the variation of manual segmentations) instead of just one. Potentially, with this approach, there is a higher chance that a clinician prefers one of automatically produced segmentation variants. This work focuses on evaluating this method on prostate segmentation in MRI scans used for brachytherapy and investigating its potential clinical usefulness.</div></div><div><h3>Materials and Methods</h3><div>Three experienced radiation oncologists graded (per-slice and per-scan) segmentations produced by our method, reference segmentations (manually created and used for brachytherapy treatment planning) and segmentations produced by a classical DL method. The study was retrospective and the way the segmentation was generated (our method, classical DL method, or manually) was blinded for the clinicians. The grades reflect the amount of manual correction required. Additionally, the clinicians were asked to rank segmentations to evaluate which one is preferred for each scan. The study was performed on 13 prostate cancer patients.</div></div><div><h3>Results</h3><div>Segmentations produced by our method are graded as requiring no manual correction in 292/576 (51 %) slices compared to 240/576 (42 %) slices in the case of the segmentations produced by a classical DL method. Furthermore, in fewer slices, 38 (6.6 %) vs. 48 (8.3 %), segmentations by our method were graded as unacceptable.</div></div><div><h3>Conclusion</h3><div>Our study has demonstrated that deep-learning-based segmentation methods can produce high-quality segmentations. Our method was evaluated better than a classical DL method, indicating the potential for integration into clinical practice.</div></div>","PeriodicalId":36850,"journal":{"name":"Physics and Imaging in Radiation Oncology","volume":"36 ","pages":"Article 100852"},"PeriodicalIF":3.3,"publicationDate":"2025-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145681030","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-10-01DOI: 10.1016/j.phro.2025.100844
Maria Thor , Aditya Apte , Milan Grkovski , Charles B. Simone II , Daphna Y. Gelblum , Masoud Zarepisheh , Puneeth Iyengar , Abraham J. Wu , Jacob Y. Shin , Tafadzwa Chaunzwa , Jennifer Ma , David Billing , Mark Dunphy , Jamie E. Chaft , Daniel R. Gomez , Joseph O. Deasy , Narek Shaverdian
Background and purpose
Early onset radiation pneumonitis (RPEarly) after concurrent chemoradiotherapy (cCRT) can lead to consolidation immunotherapy (IO) discontinuation, and poor survival in locally advanced non-small cell lung cancer (LA-NSCLC). This work assessed the external validity of a previously published RPEarly risk model.
Material and methods
The RPEarly risk model utilizes pretreatment 18F-FDG PET/CT imaging of the normal lungs and the mean lung dose (MLD). The 90th percentile of the standardized uptake value (SUVP90) and the MLD model parameters from the previous derivation cohort (N = 160) were applied in the independent cohort (50 consecutive LA-NSCLC patients treated with cCRT and IO) where model performance was evaluated (area under the receiver-operating characteristic curve (AUC), p-values, and the Hosmer-Lemeshow test (pHL)).
Results
Seven patients (14 %) developed RPEarly. Model performance of the previously developed SUVP90 and MLD model improved with re-fitting (AUC = 0.76 vs. 0.72; p = 0.01 vs. 0.10; pHL = 0.66 vs. 0.94). Above a clinically desirable 10 % predicted RPEarly, after refitting model coefficients in the combined derivation and validation cohorts (N = 210), the MLD was 13 ± 2.2 EQD23 Gy (SUVP90 = 1.2 ± 0.3) above the RPEarly risk threshold vs. 8.5 ± 2.6 EQD23 Gy (0.9 ± 0.2) below the threshold. For an SUVP90 of 1.1 and an MLD of 11 Gy EQD23 Gy, 25/27 patients developing RPEarly were captured.
Conclusion
The previously developed SUVP90 and MLD-based risk model for RPEarly demonstrated a high probability to correctly predict RPEarly in the independent cohort. This now validated RPEarly risk model with derived high-risk indications could enable personalized thoracic RT planning to reduce the risk of RPEarly and of discontinuing life-prolonging IO post-cCRT.
背景和目的同步放化疗（cCRT）后早发性放射性肺炎（RPEarly）可导致局部晚期非小细胞肺癌（LA-NSCLC）的巩固免疫治疗（IO）中断和生存率低。这项工作评估了先前发表的RPEarly风险模型的外部有效性。材料和方法RPEarly risk模型采用预处理的18F-FDG PET/CT正常肺成像和平均肺剂量（MLD）。在独立队列（50例连续接受cCRT和IO治疗的LA-NSCLC患者）中，采用标准化摄取值（SUVP90）的第90百分位和先前衍生队列（N = 160）的MLD模型参数，评估模型性能（接受者-工作特征曲线下面积（AUC）、p值和Hosmer-Lemeshow检验（pHL））。结果早期发病7例（14%）。先前开发的SUVP90和MLD模型的模型性能通过重新拟合得到改善（AUC = 0.76 vs. 0.72; p = 0.01 vs. 0.10; pHL = 0.66 vs. 0.94）。在推导和验证联合队列（N = 210）中修正模型系数后，在临床所需的10%以上预测RPEarly， MLD比RPEarly风险阈值高13±2.2 EQD23 Gy (SUVP90 = 1.2±0.3)，比阈值低8.5±2.6 EQD23 Gy（0.9±0.2）。SUVP90为1.1，MLD为11 Gy EQD23 Gy，捕获了25/27的早期发展患者。结论先前建立的基于SUVP90和mld的RPEarly风险模型在独立队列中正确预测RPEarly的概率很高。现在，这个经过验证的RPEarly风险模型及其衍生的高风险适应症可以实现个性化的胸部RT计划，以降低RPEarly的风险和ccrt后停止延长生命的IO的风险。
{"title":"Prospective validation of a pretreatment 18F-FDG PET/CT and mean lung dose model for early radiation pneumonitis","authors":"Maria Thor , Aditya Apte , Milan Grkovski , Charles B. Simone II , Daphna Y. Gelblum , Masoud Zarepisheh , Puneeth Iyengar , Abraham J. Wu , Jacob Y. Shin , Tafadzwa Chaunzwa , Jennifer Ma , David Billing , Mark Dunphy , Jamie E. Chaft , Daniel R. Gomez , Joseph O. Deasy , Narek Shaverdian","doi":"10.1016/j.phro.2025.100844","DOIUrl":"10.1016/j.phro.2025.100844","url":null,"abstract":"<div><h3>Background and purpose</h3><div>Early onset radiation pneumonitis (RP<sub>Early</sub>) after concurrent chemoradiotherapy (cCRT) can lead to consolidation immunotherapy (IO) discontinuation, and poor survival in locally advanced non-small cell lung cancer (LA-NSCLC). This work assessed the external validity of a previously published RP<sub>Early</sub> risk model.</div></div><div><h3>Material and methods</h3><div>The RP<sub>Early</sub> risk model utilizes pretreatment 18F-FDG PET/CT imaging of the normal lungs and the mean lung dose (MLD). The 90th percentile of the standardized uptake value (SUV<sub>P90</sub>) and the MLD model parameters from the previous derivation cohort (N = 160) were applied in the independent cohort (50 consecutive LA-NSCLC patients treated with cCRT and IO) where model performance was evaluated (area under the receiver-operating characteristic curve (AUC), <em>p-values</em>, and the Hosmer-Lemeshow test (<em>pHL</em>)).</div></div><div><h3>Results</h3><div>Seven patients (14 %) developed RP<sub>Early</sub>. Model performance of the previously developed SUV<sub>P90</sub> and MLD model improved with re-fitting (AUC = 0.76 <em>vs.</em> 0.72; p = 0.01 <em>vs.</em> 0.10; pHL = 0.66 <em>vs.</em> 0.94). Above a clinically desirable 10 % predicted RP<sub>Early</sub>, after refitting model coefficients in the combined derivation and validation cohorts (N = 210), the MLD was 13 ± 2.2 EQD2<sub>3</sub> Gy (SUV<sub>P90</sub> = 1.2 ± 0.3) above the RP<sub>Early</sub> risk threshold <em>vs.</em> 8.5 ± 2.6 EQD2<sub>3</sub> Gy (0.9 ± 0.2) below the threshold. For an SUV<sub>P90</sub> of 1.1 and an MLD of 11 Gy EQD2<sub>3</sub> Gy, 25/27 patients developing RP<sub>Early</sub> were captured.</div></div><div><h3>Conclusion</h3><div>The previously developed SUV<sub>P90</sub> and MLD-based risk model for RP<sub>Early</sub> demonstrated a high probability to correctly predict RP<sub>Early</sub> in the independent cohort. This now validated RP<sub>Early</sub> risk model with derived high-risk indications could enable personalized thoracic RT planning to reduce the risk of RP<sub>Early</sub> and of discontinuing life-prolonging IO post-cCRT.</div></div>","PeriodicalId":36850,"journal":{"name":"Physics and Imaging in Radiation Oncology","volume":"36 ","pages":"Article 100844"},"PeriodicalIF":3.3,"publicationDate":"2025-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145221817","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号